A Joint Use of Both PAPR Reduction and Neural Network Predistortion Approaches to Compensate the HPA Non-Linearity Impact on FBMC/OQAM Signals